On Thu, Jan 09, 2020 at 03:53:40PM +0100, wlallem...@haproxy.com wrote:
> Regarding the master CLI:
> 
> On Thu, Jan 09, 2020 at 02:31:19PM +0000, William Dauchy wrote:
> > that's a good subject to talk about. While looking at this I also
> > wondered why it was not possible to configure it in the config file; to my
> > knowledge there is no explanation about it in the doc, nor in the code,
> > but I might have missed something.
> > I did not know about those concerns around being able to access CLI even
> > if the config file is corrupted, thanks for the clarification.
> > 
> 
> In fact it's supposed to be configured in your unit file or from your init
> script (/etc/{default,sysconfig} etc.).
> 
> The problem is that the configuration for HAProxy and for the processes are in
> the same configuration file, and if something is wrong, it won't work anymore.
> And the master is supposed to survive a reload with a wrong configuration.

Even without a wrong configuration the important point is that the master
must not change its external interfaces (i.e. the socket) if the config
changes during a reload, or it really becomes a mess. The typical issue
will be that it would then need to consider rebinding across reloads,
which means closing and re-opening the socket, and that's terribly bad.
First it would prevent from being able to maintain a connection across
reloads, second it means that if anything goes wrong during this moment,
you definitely lose access to the master's socket with no option to
recover it.

Willy

Reply via email to